TW201310561A - Wafer defect analysis and trouble-shooting method of defect cause - Google Patents

Abstract

The present invention is to dispose a plurality of defect detection stations in a wafer production process and define a detection station finding out the abnormal wafer as the abnormality detection station, while setting up all the process points between the abnormality detection station and at least one previous detection station as the cause interval to further utilize the wafer defect information to carry out problem detection in terms of the common factors of each process point. By means of step-by-step selection of the common factors and aiming at the selected common factors to carry out the work area grouping towards the detected wafer, each of the process points can obtain a chi-squared value by computation and the damage rate and the defection statistical information corresponding to each work area. The present invention, by means of the chi-squared value of each process point, sorts the doubt possibility of each process point, while being matched with the damage rate and the defection statistical information corresponding to each work area to judge the problem process point, thereby quickly finding out the factor lowering the process yield to increase the production efficiency.

Description

Wafer defect analysis and method for finding defects

The invention relates to a method for finding a problem of a semiconductor process, in particular to a method for searching for wafer defects and finding a cause of defects.

As the requirements for area and efficiency of integrated circuits are increasing, semiconductor process technology is becoming more and more complex. According to the current standard, a wafer is cleaned, ground, filmed, developed, etched, etc. from the wafer. The process steps require hundreds of processes, and in the manufacturing process, it is inevitable that there will be problems with wafer defects caused by process conditions or machine table differences. If the cause of the defect cannot be detected immediately, it will often cause a problem that the process yield is greatly reduced.

In order to solve this problem, for example, the Republic of China Patent Bulletin No. 351827 "Fabric Process Yield Common Factor Analysis Method" discloses a method for effectively estimating the problem of low yield, which utilizes a plurality of wafers. The batch contains low-yield wafer lots with problems, and analyzes the common factors of the multiple wafers at each process point. Each process point can be calculated to obtain a damage rate and possibility. = count of bad lots / count of total lots, which is used to indicate the probability of a problem with the process point, so that the damage rate (with appropriate weighting) is The process points are sorted according to the possibility of the case, and the problem points in the process path can be quickly found to improve the yield. However, the conventional method has three problems, and its efficiency and accuracy in finding the problem point are limited. First, the conventional method searches for all common factors of all process points, and the range is too large; second, it causes low The factors of yield are not single, and may be caused by the accumulation of defects at each process point, or by factors such as poor electrical performance, and the conventional method is not easy to find the true cause; third, the conventional method only considers each The destruction rate of the process point is easy to ignore the misjudgment caused by the different production ratios of various machines, parts, masks and other common factors in each process point. Trying to use the production machine as the common factor to be analyzed, due to the production considerations, the production ratio of each machine is different at each process point. When the production is highly concentrated on a single machine, it is easy to cause the damage rate to be too high. Caused a false positive.

The main object of the present invention is to solve the problem that the accuracy of the conventional detection method is insufficient and there is a possibility of misjudgment.

To achieve the above object, the present invention provides a wafer defect analysis and a method for finding a defect cause by setting a plurality of detection stations in a wafer preparation process, and defining a detection station for detecting an abnormal wafer as an abnormality detection station. The method includes the following steps:

S1: selecting a defect type to be analyzed, and selecting, by computer screening or manual coding, the number of defects of the type of the plurality of wafers passing through the abnormality detecting station, and defining between the abnormal detecting station and at least the previous detecting station a factor interval having at least one process point in the range, the process point having a plurality of work areas, each of the work areas for performing a corresponding process on the plurality of wafers;

S2: determining, according to the experience and judgment of the user, a defect quantity standard value (SPEC), determining the normality and abnormality of the plurality of wafers, and marking each tested wafer as a normal wafer or an abnormal wafer;

S3: selecting a common factor, one of the plural factors of the process points in the cause interval is selected as an analysis factor, and the common factor may be a machine, a machine type, a machine model, a mask, and a furnace tube. Location and so on.

S4: matching the analysis factor with the at least one process point, calculating a Chi-square value for each process point and a NG Ratio corresponding to each work area for each of the tested wafer groups, Wherein the chi-square value represents a correlation between the abnormal wafer quantity and the analysis factor in the analysis factor, and the damage rate is the sum of the abnormal wafer quantity divided by the sum of the abnormal wafer quantity and the normal wafer quantity;

S5: if the chi-square value indicates that the correlation between the abnormal wafer quantity and the analysis factor is small, returning to step S4, selecting another common factor, if the chi-square value indicates the abnormal wafer quantity and the analysis factor If the correlation is large, continue to analyze the cause.

As can be seen from the above description, the present invention can be divided into three main stages: information input, information analysis, and information output. When inputting information, first select the specific defect type to be analyzed, and set the filtering conditions to obtain a single defect type by using defect information on the plurality of wafers such as defect size, distribution on the wafer, manual coding, and the like. The number of wafer defects is then input into the analysis system, and the value of the standard number of defects (SPEC) is set to classify the plurality of wafers passing through the anomaly detection station into normal wafers and abnormal wafers. Then, in the information analysis stage, select the common factors of the process factors such as model, machine, mask and tube position, use the chi-square value to sort the possible process points in the cause interval, and finally output the suspected process. The card's square value and complex number correspond to the damage rate of the work area for subsequent analysis.

Compared with the prior art, the advantages of the present invention are:

1. Use the chi-square value to determine the correlation between the abnormal wafer and the analysis factor, thereby avoiding the misjudgment caused by the different production factors due to the common factors on the same process point.

2. Since a plurality of defect detection stations are set up in the wafer preparation process, the anomaly detection station and the previous normal detection station can be defined as the cause interval, and the process is greatly reduced compared with the conventional method. Its search range.

3. According to the type of single defect, the cause is single, avoiding the result of analysis of different types of defects and analyzing the misjudgment.

The detailed description and technical contents of the present invention will now be described as follows:

Referring to FIG. 1 and FIG. 2, the present invention relates to wafer defect analysis and defect cause finding method, which is to set a plurality of detection stations in a wafer preparation process and define an abnormal wafer. The detection station is an anomaly detection station, and the method includes the following steps:

S1: filtering and screening specific types of defect information, selecting one type of defect to be analyzed, which is selected by computer or manually coded to obtain the number of defects of the type of the plurality of wafers passing through the abnormality detecting station, because different process steps It may cause different defects such as cracking, incomplete exposure, scratches, potholes, bulging and smudging, etc., using computer screening or manual coding to filter out a type of defect to be analyzed. Further, defining a cause interval 1 between the abnormality detecting station and at least the previous detecting station, and obtaining the number of defects on the plurality of wafers passing through the abnormal detecting station, in the embodiment, the first The detection station 10 and a second detection station 20 are exemplified. The second detection station 20 is assumed to be the abnormality detection station, and the first detection station 10 is a normal detection station whose inspection result is normal. Therefore, the first detection is performed. The cause interval 1 is between the station 10 and the second detection station 20. Please refer to FIG. 3 for the cause interval 1 having at least one process point. The process point has three in this embodiment. The first process point 31, the second process point 32, and the third process point 33, respectively, wherein each process point 31, 32, 33 can be grouped into a plurality of work areas corresponding to the analysis factors, and each of the work areas is used for For the corresponding process of the plurality of wafers, in the embodiment, the first process point 31 has four working areas A1, A2, A3, and A4; the second process point 32 has three working areas B1, B2. B3; and the third process point 33 has three work areas C1, C2, C3.

S2: classification of normal wafers and abnormal wafers, according to a defect quantity standard value (SPEC), a normal wafer and an abnormal wafer are classified into a plurality of wafers to obtain a normal wafer number and an abnormal crystal. The number of circles. For example, the user can define the standard value of the single defect quantity according to experience. If the order is 50, it means that if there are more than 50 such defects in a wafer, it is classified as an abnormal wafer, if less than 50. It is classified as a normal wafer, so that the number of normal wafers and the number of abnormal wafers can be obtained by classifying a plurality of wafers passing through the abnormality detecting station.

S3: selecting a common factor, one of the plurality of common factors of the process point in the cause interval 1 is selected as an analysis factor, and the common factor may be a machine, a machine type, a mask, a tube position, and a wash. The single-batch/two-batch inbound of the net machine, the use cycle of the machine parts, the process recipe and the wafer waiting time, etc., can refer to the production information shared by the wafers at each process point, taking "Figure 3" as an example. The machine is used as an analysis factor. Since each process point represents its corresponding work, such as yellow light development, etching, film, etc., each process point can have multiple machines, that is, In the work area referred to in the invention, the wafers passing through the respective process points can be respectively arranged in parallel by the work areas. In other words, in the embodiment, the work areas A1, A2, A3, and A4 respectively represent a set of work machines in the first process point 31, and the work areas B1, B2, and B3 respectively represent a group. The working machine in the second process point 32; the working areas C1, C2, and C3 respectively represent a group of working machines in the third process point 33.

S4: Calculate a Chi-square value of each of the process points 31, 32, and 33, and a NG Ratio corresponding to the plurality of work areas, and calculate the card by using the at least one process point with the analysis factor. a square value and the damage rate, wherein the chi-square value represents a correlation between the abnormal wafer quantity and the analysis factor in the analysis factor, and the damage rate is the number of the abnormal wafers in each working area divided by the abnormal crystal The sum of the number of circles and the number of normal wafers. The number of wafers that work in each work area is not necessarily equal. The main reason is that in some cases, it is necessary to select a process that has one of the same process points to bear a heavier process. One of the work areas has a larger number of wafers than the rest of the work area. Also, because each work area is operated by an independent machine, each has its corresponding damage rate. In addition, each of the working areas has an abnormal wafer number actual value Nx and an abnormal wafer quantity expected value Ex, which is obtained by obtaining the sum of (Nx-Ex) ² /Ex of each working area. If the total value of (Nx-Ex) ² /Ex is larger, the chi-square value obtained by looking up the table will be closer to 0, indicating that the correlation is large. On the contrary, if (Nx-Ex) ² The smaller the value of /Ex is, the closer the chi-square value obtained by the look-up table is to 1, indicating that the correlation is small, and the chi-square value is not greater than 1, wherein the abnormal wafer expected value Ex is assumed to be the abnormal wafer. Regardless of the analysis factor, it is set according to the number of wafers in which the work area performs the process work, that is, if the number of wafers in the work area is large, the preset value of the abnormal wafer will be large. By this, it can be known whether there is a significant difference between the number of abnormal wafers in each process point and the expected number of abnormal wafers, so as to eliminate blind spots caused by only looking at the number of abnormal wafers.

S5: using the chi-square value to analyze and judge the damage rate, and matching the defect statistical information or the defect trend map with other relevant information for analysis and judgment, if the chi-square value indicates that the abnormal wafer quantity is related to the analysis factor If the sex is small, return to step S4 and select another common factor, that is, the chi-square value approaches 1 , which means that the selected analysis factor is not a factor mainly causing the defect, and thus the common factor needs to be re-selected as the analysis factor. If the chi-square value indicates that the number of abnormal wafers is highly correlated with the analysis factor, that is, the chi-square value approaches 0, indicating that the selected analysis factor is the main cause in the cause interval 1 Then, the cause analysis is continued, and if there is only one process point whose chi-square value approaches zero, the source of the cause can be judged according to the damage rate of the work area in the process point.

If the chi-square value of a plurality of process points indicates that the number of abnormal wafers is highly correlated with the analysis factor, the following steps are performed:

S6: performing cross-comparison of a plurality of process points, because of the two process points in the front and the back, it is possible that the previous process point causes a specific defect in the work process due to the work area, and the wafer passing through the work area of the cause Due to factors such as coincidence or production configuration, it passes through a certain working area of the latter process point. Therefore, the result of the chi-square value and the damage rate cannot be determined by the process point. Therefore, if the chi-square value is Process points that approach zero have multiple, and must be cross-checked to eliminate the effects of potential factors. The process point shown in FIG. 3 that the chi-square value approaches 0 has a first process point 31 and a second process point 32. Therefore, the intersection of the first process point 31 and the second process point 32 is required. For comparison, please refer to "Figure 4" and "Figure 5". Figure 4 shows the average, median, maximum, damage rate and working wafer of abnormal wafers corresponding to each working area of a single process point. The quantity and the two process points are matched for cross-analysis. As shown in Figure 5, it is clear that the work areas with high damage rate include B1 and B2, so work areas B1 and B2 can be used. Judging as an abnormal cause.

Using the chi-square value for defect analysis and judgment can improve the accuracy. The reason is that the chi-square value is based on a hypothesis. First, if the abnormality of the tested wafer is not related to the common factor to be tested, the abnormal wafer is tested. The distribution should be in accordance with the production ratio of the common factor, that is, the number of abnormal wafers contained in each group (classified by the factors to be analyzed) should be proportional to the number of wafers in each group, if the actual number of abnormal wafers conforms to each group The production ratio can be calculated as a chi-square value close to one, indicating that the abnormal wafer has nothing to do with the factor to be analyzed, and vice versa. Therefore, the conventional analysis method can be excluded, and only the destruction rate is ignored and the production ratio is neglected. Misjudgment.

In summary, the advantages of the present invention over conventional techniques are:

1. Use the chi-square value to determine the correlation between the abnormal wafer and the analysis factor, thereby avoiding the misjudgment caused by the common production factors (such as machine, model, mask, etc.) due to different production quantities at the same process point.

2. Since a plurality of defect detection stations are set up in the wafer preparation process, the anomaly detection station and the previous normal detection station can be defined as the cause interval, and the process is greatly reduced compared with the conventional method. Its search range.

3. Analyze according to the type of single defect, avoid the result of analysis of different types of defect interference and analyze the problem of misjudgment.

4. Cross-comparison is used to avoid misjudgment caused by potential factors when there is a process point where the complex chi-square value approaches zero.

Therefore, the present invention is highly progressive and conforms to the requirements of the invention patent application, and the application is filed according to law, and the praying office grants the patent as soon as possible.

The present invention has been described in detail above, but the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the scope of the invention. That is, the equivalent changes and modifications made by the scope of the present application should remain within the scope of the patent of the present invention.

S1~S6. . . step

1. . . Cain interval

10. . . First test station

20. . . Second test station

31. . . First process point

32. . . Second process point

33. . . Third process point

FIG. 1 is a schematic diagram of a process path according to a preferred embodiment of the present invention.

2 is a flow chart showing the steps of a preferred embodiment of the present invention.

FIG. 3 is a schematic diagram of chi-square value analysis according to a preferred embodiment of the present invention.

4 is a schematic diagram of single process point analysis according to a preferred embodiment of the present invention.

FIG. 5 is a schematic diagram of cross-comparison analysis according to a preferred embodiment of the present invention.

S1~S6. . . step

Claims (7)

A wafer defect analysis and a method for finding a defect cause a plurality of detection stations in a wafer preparation process, and defining a detection station for detecting an abnormal wafer as an abnormality detection station, the method comprising the following steps:
S1: selecting a defect type to be analyzed, and defining a cause interval between the abnormality detecting station and at least the previous detecting station, and obtaining the number of defects on the plurality of wafers passing through the abnormal detecting station, the cause interval Having at least one process point, the process point has a plurality of work areas, each of the work areas for performing corresponding process work on the plurality of wafers;
S2: classifying a normal wafer and an abnormal wafer according to a defect quantity standard value to obtain a normal wafer quantity and an abnormal wafer quantity;
S3: selecting a common factor, and selecting one of the plural factors of the process points in the cause interval as an analysis factor;
S4: using the at least one process point to cooperate with the analysis factor, calculating a card square value and a plurality of damage rates corresponding to the working areas, wherein the card value represents the number of the abnormal wafers and the analysis factor Correlation, and the damage rate is the sum of the abnormal wafers divided by the sum of the abnormal wafer number and the normal wafer number;
S5: if the chi-square value indicates that the correlation between the abnormal wafer quantity and the analysis factor is small, returning to step S4, selecting another common factor, if the chi-square value indicates the abnormal wafer quantity and the analysis factor If the correlation is large, continue to analyze the cause. For example, the wafer defect analysis and the method for finding the cause of defects described in the first application of the patent scope, wherein the cause interval is the interval between the abnormality detecting station and the previous normal detecting station. The method for searching for wafer defects and the method for finding the cause of defects, as described in claim 1, wherein in step S3, the common factor is selected from the group consisting of a machine, a machine type, a mask, and a furnace tube. The group that makes up. For example, the wafer defect analysis and the defect cause finding method described in claim 1 wherein if the chi-square value of the plurality of process points indicates that the abnormal wafer quantity has a large correlation with the analysis factor, then one is performed. Step S6: performing cross-comparison of a plurality of process points, and determining a work area including a high damage rate as an abnormal cause. The method for searching for wafer defects and the method for finding the cause of defects, as described in claim 1, wherein in step S4, each of the working areas has an abnormal wafer actual value Nx and an abnormal wafer expected value Ex, The chi-square value is obtained by looking up the table after obtaining the sum of (Nx-Ex) ² /Ex of each work area, and if the value of (Nx-Ex) ² /Ex is larger, the chi-square value obtained by looking up the table is obtained. The closer to 0, the greater the correlation. On the contrary, if the value of (Nx-Ex) ² /Ex is smaller, the chi-square value obtained by looking up the table is closer to 1, indicating that the correlation is small, and the chi-square The value is not greater than 1. For example, the wafer defect analysis and the defect cause finding method described in claim 5, wherein the abnormal wafer expected value is set according to the number of wafers in which the work area performs the process work. For example, the wafer defect analysis and the method for finding the cause of defects described in claim 1 are further determined in step S5 in conjunction with the defect statistical information or the defect trend graph.